Extremal length and the modulus of curve families

Intuition Beginner

Imagine a flat region of the plane and a whole family of routes that cross it from one side to the other. Extremal length is a single number that measures how hard it is, on the whole, to get any route across. A wide and short channel is easy to cross, and gets a small extremal length. A long and thin corridor is hard to cross, and gets a large extremal length. The number rises with difficulty.

What makes this number special is that it ignores the artist who drew the region and keeps only its true shape. If you take the region and bend it with a conformal map — a map that preserves angles and small circles — the picture changes, the routes change, and yet the extremal length reads exactly the same. It is a ruler that bends along with the region but never changes its readings.

A useful picture is a river delta. The water has many channels to flow through. Where the delta is wide and the channels are short and fat, water crosses with ease, and the crossing has small extremal length. Where the delta narrows into one long thin run, water struggles, and the extremal length is large. The modulus is just the same information read upside down: it is one divided by the extremal length, so it is large where crossing is easy and small where crossing is hard.

Visual Beginner

A schematic of an annular region (a disc with a smaller disc removed from its centre) with several curves drawn going around the central hole, like rings on a tree stump. Beside it the same region is shown after a conformal map has straightened it into a rectangle, with the rings now horizontal lines and the radial direction the height of the rectangle. An arrow between the two pictures is labelled "same modulus": both pictures encode the one number $\log (R/r)$ that measures how thick the ring is.

Worked example Beginner

Take a flat rectangle that is $3$ units wide and $1$ unit tall. Consider the family of routes that cross it from the left edge to the right edge. The honest way to measure these routes is to use the rectangle's own flat ruler, with the metric weight set to $1$ everywhere.

The shortest such route is a straight horizontal segment, with length $3$. The area of the rectangle is width times height, which is $3\times 1=3$. The extremal length of the crossing family is the shortest length squared, divided by the area: $3\times 3=9$, divided by $3$, which is $3$. So the left-to-right extremal length equals $3$.

Now try the routes that go the other way, from the bottom edge to the top edge. The shortest is a vertical segment of length $1$, the area is still $3$, and the extremal length is $1\times 1=1$ divided by $3$, which is $1/3$.

What this tells us: the long way across (left to right, through the width $3$) is hard and scores $3$; the short way across (bottom to top, through the height $1$) is easy and scores $1/3$. The two numbers multiply to $1$. A rectangle's two crossing problems are reciprocals of each other, and the number $3$ is exactly the rectangle's aspect ratio.

Check your understanding Beginner

Formal definition Intermediate+

Let $\Omega \subseteq \mathbb{C}$ be a plane region and let $\Gamma$ be a family of (locally rectifiable) curves in $\Omega$. A conformal metric on $\Omega$ is a Borel-measurable function $\rho :\Omega \to [0,\infty ]$, interpreted as the length element $\rho |dz|$. For such a $\rho$ define the $\rho$-length of the family and the $\rho$-area by

$$L_{\rho }(\Gamma )=\underset{\gamma \in \Gamma }{\inf }{\int }_{\gamma }\rho |dz|,A_{\rho }={\iint }_{\Omega }{\rho }^{2}dxdy.$$

The metric $\rho$ is admissible when $0<A_{\rho }<\infty$. The extremal length of $\Gamma$ is the conformally invariant ratio

$$\lambda (\Gamma )=\underset{\rho \mathrm{admissible}}{\sup }\frac{L_{\rho }(\Gamma {)}^2}{A_{\rho }},$$

and the modulus is its reciprocal $\mathrm{mod}(\Gamma )=1/\lambda (\Gamma )$. The quotient $L_{\rho }^2/A_{\rho }$ is unchanged when $\rho$ is multiplied by a positive constant, so one may normalise either $L_{\rho }(\Gamma )=1$ or $A_{\rho }=1$ without affecting the supremum. The definition follows Ahlfors-Sario Ch. II ^{[Ahlfors and Sario Riemann Surfaces]}.

The construction follows the originating treatment of Ahlfors and Beurling, who isolated extremal length as the first systematically conformally invariant length-area functional ^{[Ahlfors and Beurling 1950]}. Two ordering relations organise the calculus of curve families. Write ${\Gamma }_1<{\Gamma }_2$, and say ${\Gamma }_2$ overflows ${\Gamma }_1$, when every curve ${\gamma }_2\in {\Gamma }_2$ contains a curve ${\gamma }_1\in {\Gamma }_1$ as a subcurve. The families $\Gamma ={\bigcup }_k{\Gamma }_k$ are combined in series when each $\gamma \in \Gamma$ overflows one curve from every ${\Gamma }_k$ supported in disjoint subregions, and in parallel when the ${\Gamma }_k$ live in disjoint subregions and each $\gamma \in \Gamma$ lies in a single ${\Gamma }_k$.

Counterexamples to common slips

• The supremum is over all admissible $\rho$, not over the flat metric alone; using only $\rho =1$ gives a lower bound for $\lambda (\Gamma )$, never the value. The extremal-length number is attained by a specific extremal metric that is rarely constant.
• $L_{\rho }(\Gamma )$ is an infimum over the family before the supremum over metrics, so enlarging $\Gamma$ (adding more, possibly shorter, curves) can only decrease $L_{\rho }(\Gamma )$ and hence decrease $\lambda (\Gamma )$. A larger family is easier to cross, not harder.
• The modulus and the extremal length are reciprocals, so the qualitative readings are opposite: easy-to-cross families have small $\lambda$ and large $\mathrm{mod}$. Mixing the two conventions inverts every inequality.

Key theorem with proof Intermediate+

Theorem (conformal invariance of extremal length). Let $f:\Omega \to {\Omega }^{\prime }$ be a conformal (biholomorphic) map, and let $\Gamma$ be a curve family in $\Omega$ with image family $f(\Gamma )=\{f\circ \gamma :\gamma \in \Gamma \}$ in ${\Omega }^{\prime }$. Then $\lambda (f(\Gamma ))=\lambda (\Gamma )$.

Proof. Let ${\rho }^{\prime }$ be any admissible metric on ${\Omega }^{\prime }$. Define a metric $\rho$ on $\Omega$ by the pullback rule

$$\rho (z)={\rho }^{\prime }(f(z))|f^{\prime }(z)|.$$

For a curve $\gamma \in \Gamma$ with image ${\gamma }^{\prime }=f\circ \gamma$, the change-of-variables formula for the line integral gives

$${\int }_{{\gamma }^{\prime }}{\rho }^{\prime }|dw|={\int }_{\gamma }{\rho }^{\prime }(f(z))|f^{\prime }(z)||dz|={\int }_{\gamma }\rho |dz|,$$

so the two metrics assign equal lengths to corresponding curves, and therefore $L_{{\rho }^{\prime }}(f(\Gamma ))=L_{\rho }(\Gamma )$. For the areas, the Jacobian of the conformal map $f$ is $|f^{\prime }(z){|}^2$, so

$$A_{{\rho }^{\prime }}={\iint }_{{\Omega }^{\prime }}{\rho }^{\prime }(w{)}^{2}dudv={\iint }_{\Omega }{\rho }^{\prime }(f(z){)}^2|f^{\prime }(z){|}^{2}dxdy={\iint }_{\Omega }\rho (z{)}^{2}dxdy=A_{\rho }.$$

Hence $L_{{\rho }^{\prime }}(f(\Gamma ){)}^2/A_{{\rho }^{\prime }}=L_{\rho }(\Gamma {)}^2/A_{\rho }$. The pullback $\rho \mapsto {\rho }^{\prime }$ is a bijection between admissible metrics on the two regions (its inverse is the pullback along $f^{-1}$, using $(f^{-1}{)}^{\prime }(w)=1/f^{\prime }(f^{-1}(w))$), and it preserves the ratio term by term. Taking the supremum over admissible metrics on each side, the two suprema agree: $\lambda (f(\Gamma ))=\lambda (\Gamma )$. $\square$

Bridge. Conformal invariance is the foundational reason extremal length earns the name conformal invariant and builds toward the type problem for open Riemann surfaces, where the prerequisite exhaustion machinery of 06.11.03 feeds a divergent extremal-length series into a parabolicity criterion. The same pullback computation appears again in the proof that the modulus of a ring domain is the unique conformal invariant of an annulus, which is exactly the statement that biholomorphic annuli have equal radius ratios. The length-area exchange under the Jacobian $|f^{\prime }{|}^2$ generalises to the modulus of a quadrilateral, the quasiconformal distortion bound, and ultimately the Teichmüller extremal problem, where the extremal length of a measured foliation becomes a coordinate on Teichmüller space. Putting these together, the bridge is that a single length-area ratio, invariant under the maps of 06.01.06, simultaneously measures crossing difficulty, classifies ring domains, and detects the type of a surface; this is exactly the unifying role Ahlfors and Beurling assigned to the invariant when they introduced it.

Exercises Intermediate+

Lean formalization Intermediate+

Mathlib does not currently name conformal metrics on a plane region, the extremal-length functional, or the modulus of a ring domain. A proposed signature, in Lean 4 / Mathlib syntax, sketching the conformal-invariance target:

-- Sketch only; no current Mathlib coverage. See lean_mathlib_gap.
import Mathlib.MeasureTheory.Integral.Bochner
import Mathlib.Analysis.SpecialFunctions.Complex.Analytic

namespace Codex.RiemannSurfaces.OpenSurfaces

-- A conformal metric is a Borel-measurable density ρ : Ω → [0,∞].
-- L ρ Γ is the family-infimum of line integrals; A ρ the area integral.
variable {Ω Ω' : Set ℂ}

noncomputable def extremalLength
    (Γ : Set (ℝ → ℂ)) (Lfun : (ℂ → ENNReal) → ENNReal)
    (Afun : (ℂ → ENNReal) → ENNReal) : ENNReal :=
  ⨆ ρ, (Lfun ρ) ^ 2 / (Afun ρ)

-- Conformal invariance: a biholomorphism f : Ω → Ω' leaves extremal length fixed.
theorem extremalLength_conformal_invariant
    (f : ℂ → ℂ) (hf : Set.InjOn f Ω)
    (Γ : Set (ℝ → ℂ)) :
    extremalLength (fun γ => f ∘ γ '' Γ) LfunImage AfunImage
      = extremalLength Γ Lfun Afun := by
  sorry

end Codex.RiemannSurfaces.OpenSurfaces

The proof depends on names that do not currently exist in Mathlib: the line-integral length functional along rectifiable curves against a measurable density, the area integral as a Lebesgue integral, the family-infimum, and the change-of-variables identity ${\int }_{f\circ \gamma }{\rho }^{\prime }|dw|={\int }_{\gamma }({\rho }^{\prime }\circ f)|f^{\prime }||dz|$ together with the Jacobian area transform $|f^{\prime }{|}^2$. Each is a candidate Mathlib contribution; until then this unit ships with lean_status: none.

Advanced results Master

The extremal length attains its supremum at a distinguished extremal metric, and on a ring or quadrilateral that metric is the flat metric of the conformal rectangle model. For an admissible metric $\rho$ and the separating family $\Gamma$ of an annulus, the length-area method bounds $\lambda (\Gamma )$ above by exhibiting a single competitor and below by Cauchy-Schwarz against the area; equality pins $\rho$ to a constant multiple of $1/|z|$ almost everywhere, and the extremal metric is unique up to scale. This uniqueness is the reason the modulus is a complete conformal invariant of a ring domain: two annuli are biholomorphic if and only if their radius ratios, equivalently their moduli $2\pi /\log (R/r)$, coincide ^{[Ahlfors and Sario Riemann Surfaces]}.

Modulus of a quadrilateral. A quadrilateral is a Jordan domain $Q$ with four marked boundary points, and the Riemann map of 06.01.06 carries $(Q,\text{four points})$ onto a Euclidean rectangle $[0,m]\times [0,1]$ with the marked points at the corners. The number $m$ is the modulus of the quadrilateral, and it equals the extremal length of the family of curves joining the two sides that map to the vertical edges. Because the rectangle model is determined up to similarity, $m$ is a conformal invariant, and it is the planar archetype of every later modulus: the modulus of a ring is the cylindrical quotient of the modulus of a quadrilateral cut along a crosscut.

Composition laws as a calculus. The series inequality $\lambda (\Gamma )\ge {\sum }_k\lambda ({\Gamma }_k)$ for families overflowing disjoint subregions in series, and the parallel inequality $1/\lambda (\Gamma )\ge {\sum }_{k}1/\lambda ({\Gamma }_k)$ for families confined to disjoint subregions, are dual reformulations of one length-area estimate, with equality precisely when the extremal metrics of the pieces agree on overlaps. Combined with the overflowing/comparison principle ${\Gamma }_1<{\Gamma }_2\Rightarrow \lambda ({\Gamma }_1)\le \lambda ({\Gamma }_2)$, these laws make extremal length a genuine calculus: one bounds an unknown modulus by decomposing a region into rings and quadrilaterals whose moduli are explicit, then composing in series and parallel. The Grötzsch inequality and the Teichmüller modulus estimate are the two extremal cases of this calculus, bounding the modulus of a ring whose complementary continua are pinched to a slit or a point.

Type criterion for open surfaces. On an open Riemann surface $X$ with a bordered exhaustion from 06.11.03, let ${\Gamma }_n$ be the family separating $\partial {\Omega }_n$ from $\partial {\Omega }_{n+1}$. The surface is parabolic (its ideal boundary has zero capacity, equivalently $X$ carries no nonconstant negative subharmonic function and lies in the class $O_G$) precisely when the series ${\sum }_n\lambda ({\Gamma }_n)$ diverges, placing the ideal boundary at infinite extremal distance from any interior point. Convergence places it at finite distance and forces hyperbolicity, with a nonconstant bounded harmonic function and a Green's function. This is the extremal-length form of the type problem, and it converts a global analytic dichotomy into a single series whose terms are computable ring moduli.

Synthesis. Extremal length is the foundational reason a length-area ratio can serve as both a metric and a classifier: the conformal-invariance theorem is exactly the statement that the ratio sees only the conformal class, and putting these together with the series and parallel laws yields a calculus in which the modulus of any region is built from ring and quadrilateral moduli. The central insight is that the modulus of the round annulus, $\log (R/r)/(2\pi )$, is simultaneously the unique invariant of a ring domain, the cylindrical model behind the modulus of a quadrilateral, and the per-ring term in the parabolicity series — three readings of one extremal metric $\rho =1/|z|$. The type criterion generalises the planar reciprocal relation ${\lambda }_{\mathrm{rad}}{\lambda }_{\mathrm{sep}}=1$ to the ideal boundary of an open surface, and is dual to the capacity dichotomy of 06.11.03; what the harmonic-measure side calls zero capacity, the extremal-length side calls a divergent modulus series. This is exactly the unification Ahlfors and Beurling intended when they made conformal invariants the organising language for function-theoretic null sets, and it builds toward the Teichmüller theory of moduli, where extremal lengths of measured foliations coordinate the space of complex structures.

Full proof set Master

Proposition (modulus of the round annulus). For the annulus $\Omega =\{r<|z|<R\}$ and the family $\Gamma$ of curves separating the two boundary circles, the extremal length is $\lambda (\Gamma )=2\pi /\log (R/r)$, equivalently $\mathrm{mod}(\Gamma )=\log (R/r)/(2\pi )$. The supremum is attained, uniquely up to scale, by the radial metric ${\rho }_0=1/|z|$.

Proof. First a lower bound for $\lambda (\Gamma )$. Take the competitor ${\rho }_0(z)=1/|z|$. Every separating curve $\gamma$ winds once around the origin, so writing $z=s{e}^{i\theta }$ and parametrising, its ${\rho }_0$-length is

$${\int }_{\gamma }\frac{|dz|}{|z|}\ge \left|{\int }_{\gamma }\frac{dz}{z}\right|=\left|{\int }_{\gamma }d(\log z)\right|=2\pi ,$$

since the imaginary part of ${\int }_{\gamma }d\log z$ is the change in argument, which is $2\pi$ for a curve separating the circles. Hence $L_{{\rho }_0}(\Gamma )\ge 2\pi$, and the round circle $|z|=\sqrt{rR}$ attains it, so $L_{{\rho }_0}(\Gamma )=2\pi$. The area is

$$A_{{\rho }_0}={\iint }_{\Omega }\frac{dxdy}{|z{|}^2}={\int }_0^{2\pi }{\int }_r^R\frac{1}{s^2}sdsd\theta =2\pi {\int }_r^R\frac{ds}{s}=2\pi \log \frac{R}{r}.$$

Therefore $\lambda (\Gamma )\ge L_{{\rho }_0}(\Gamma {)}^2/A_{{\rho }_0}=(2\pi {)}^2/(2\pi \log (R/r))=2\pi /\log (R/r)$.

For the matching upper bound, let $\rho$ be any admissible metric, normalised so that $L_{\rho }(\Gamma )\ge 1$; concretely require ${\int }_{|z|=s}\rho |dz|\ge 1$ for almost every radius $s\in (r,R)$, which holds after normalisation because each circle $|z|=s$ is a separating curve. By the Cauchy-Schwarz inequality applied on the circle of radius $s$, with $|dz|=sd\theta$,

$$1\le {\left({\int }_0^{2\pi }\rho (s{e}^{i\theta })sd\theta \right)}^2\le 2\pi {\int }_0^{2\pi }\rho (s{e}^{i\theta }{)}^2{s}^{2}d\theta .$$

Dividing by $s$ and integrating over $s\in (r,R)$,

$${\int }_r^R\frac{ds}{s}\le 2\pi {\int }_r^R{\int }_0^{2\pi }{\rho }^{2}sd\theta ds=2\pi A_{\rho },$$

so $\log (R/r)\le 2\pi A_{\rho }$, that is $A_{\rho }\ge \log (R/r)/(2\pi )$. Then $L_{\rho }(\Gamma {)}^2/A_{\rho }\le 1/A_{\rho }\le 2\pi /\log (R/r)$. Taking the supremum over $\rho$ gives $\lambda (\Gamma )\le 2\pi /\log (R/r)$, matching the lower bound, so $\lambda (\Gamma )=2\pi /\log (R/r)$. Equality in Cauchy-Schwarz forces $\rho (s{e}^{i\theta })$ to be independent of $\theta$ for almost every $s$, and the normalisation then forces $\rho (z)=c/|z|$ for a constant $c>0$, establishing uniqueness of the extremal metric up to scale. $\square$

Proposition (conformal invariance). Stated and proved in the Intermediate-tier Key theorem section; the pullback $\rho (z)={\rho }^{\prime }(f(z))|f^{\prime }(z)|$ preserves both $L_{\rho }$ termwise and $A_{\rho }$ via the Jacobian $|f^{\prime }{|}^2$, and is a bijection on admissible metrics, so the suprema agree.

Proposition (series law). If $\Gamma$ overflows families ${\Gamma }_1,\dots ,{\Gamma }_n$ in disjoint subregions arranged in series, then $\lambda (\Gamma )\ge {\sum }_k\lambda ({\Gamma }_k)$.

Proof. For each $k$ choose an admissible ${\rho }_k$ on the $k$-th subregion realising $\lambda ({\Gamma }_k)$ within $\epsilon$, normalised so $L_{{\rho }_k}({\Gamma }_k)=\lambda ({\Gamma }_k)$ and hence $A_{{\rho }_k}=\lambda ({\Gamma }_k)$ (rescale ${\rho }_k$ by $\lambda ({\Gamma }_k)/L_{{\rho }_k}({\Gamma }_k)$, which multiplies $L$ by that factor and $A$ by its square). Define $\rho$ to equal ${\rho }_k$ on the $k$-th subregion and $0$ elsewhere. Each $\gamma \in \Gamma$ overflows one curve from each ${\Gamma }_k$, so ${\int }_{\gamma }\rho |dz|\ge {\sum }_k{L}_{{\rho }_k}({\Gamma }_k)={\sum }_k\lambda ({\Gamma }_k)$, giving $L_{\rho }(\Gamma )\ge {\sum }_k\lambda ({\Gamma }_k)$. The subregions are disjoint, so $A_{\rho }={\sum }_k{A}_{{\rho }_k}={\sum }_k\lambda ({\Gamma }_k)$. Therefore $\lambda (\Gamma )\ge L_{\rho }(\Gamma {)}^2/A_{\rho }\ge ({\sum }_k\lambda ({\Gamma }_k){)}^2/{\sum }_k\lambda ({\Gamma }_k)={\sum }_k\lambda ({\Gamma }_k)$. $\square$

Proposition (parallel law). If ${\Gamma }_1,\dots ,{\Gamma }_n$ live in disjoint subregions and $\Gamma ={\bigcup }_k{\Gamma }_k$, then $1/\lambda (\Gamma )\ge {\sum }_{k}1/\lambda ({\Gamma }_k)$, equivalently $\mathrm{mod}(\Gamma )\ge {\sum }_k\mathrm{mod}({\Gamma }_k)$.

Proof. Apply the series law to the dual families. Let $\rho$ be admissible for $\Gamma$ and let ${\rho }_k$ be its restriction to the $k$-th subregion. Since every $\gamma \in \Gamma$ lies in a single subregion, $L_{\rho }(\Gamma )={\min }_k{L}_{{\rho }_k}({\Gamma }_k)$, while $A_{\rho }={\sum }_k{A}_{{\rho }_k}$ by disjointness. Normalising $L_{{\rho }_k}({\Gamma }_k)=1$ for all $k$ makes $L_{\rho }(\Gamma )=1$ and $A_{{\rho }_k}\ge 1/\lambda ({\Gamma }_k)$ by definition of extremal length, so $A_{\rho }={\sum }_k{A}_{{\rho }_k}\ge {\sum }_{k}1/\lambda ({\Gamma }_k)$. Choosing ${\rho }_k$ near-extremal makes $A_{\rho }$ approach ${\sum }_{k}1/\lambda ({\Gamma }_k)$, so $1/\lambda (\Gamma )={\sup }_{\rho }{L}_{\rho }(\Gamma {)}^2/A_{\rho }$ has infimal area ${\sum }_{k}1/\lambda ({\Gamma }_k)$, giving $1/\lambda (\Gamma )\ge {\sum }_{k}1/\lambda ({\Gamma }_k)$. $\square$

Proposition (overflowing/comparison principle). If ${\Gamma }_1<{\Gamma }_2$, then $\lambda ({\Gamma }_1)\le \lambda ({\Gamma }_2)$.

Proof. Fix any admissible $\rho$. Each ${\gamma }_2\in {\Gamma }_2$ contains a subcurve ${\gamma }_1\in {\Gamma }_1$, and the integrand $\rho |dz|\ge 0$, so ${\int }_{{\gamma }_2}\rho |dz|\ge {\int }_{{\gamma }_1}\rho |dz|\ge L_{\rho }({\Gamma }_1)$. Taking the infimum over ${\gamma }_2$ gives $L_{\rho }({\Gamma }_2)\ge L_{\rho }({\Gamma }_1)$, hence $L_{\rho }({\Gamma }_2{)}^2/A_{\rho }\ge L_{\rho }({\Gamma }_1{)}^2/A_{\rho }$. Taking the supremum over $\rho$ yields $\lambda ({\Gamma }_2)\ge \lambda ({\Gamma }_1)$. $\square$

Connections Master

• Ideal boundary and exhaustions of an open Riemann surface 06.11.03. This unit supplies the conformal invariant that turns the exhaustion machinery into a type criterion: the extremal length of the family separating an interior boundary from the ideal boundary is the sum, in series, of the ring moduli of the exhaustion. Divergence of that series is the extremal-length form of the zero-capacity condition that 06.11.03 attaches to a parabolic ideal boundary, so the two units describe the same dichotomy through capacity and through modulus respectively.

• Riemann mapping theorem 06.01.06. Conformal invariance of extremal length is exactly the statement that the Riemann map of 06.01.06 preserves the modulus; the rectangle model of a quadrilateral and the cylinder model of a ring are both produced by the Riemann/uniformisation map, and the resulting moduli are the conformal invariants those maps cannot change. The uniqueness clause of the mapping theorem is what makes the modulus a complete invariant of a ring domain.

• Uniformization theorem 06.03.03. The parabolic/hyperbolic dichotomy detected by the extremal-length series is the open-surface refinement of the uniformisation trichotomy: a simply connected open surface is conformally $\mathbb{C}$ (parabolic) or $\mathbb{H}$ (hyperbolic) exactly as the modulus series of 06.11.03 diverges or converges, so extremal length is the quantitative diagnostic behind the uniformisation type.

• Moduli of Riemann surfaces 06.08.03. Extremal lengths of simple closed curves and measured foliations are coordinates on Teichmüller space, and the Teichmüller extremal problem — minimising the maximal conformal distortion in an isotopy class — is solved by the metric that makes a measured foliation's extremal length minimal. The modulus calculus of this unit is the planar germ of the moduli-space coordinates studied there.

Historical & philosophical context Master

The method of length and area predates its conformal-invariant formulation: it appears in the work of Grötzsch on extremal quasiconformal mappings in the late 1920s and in Teichmüller's 1930s extremal problems, where the modulus of a ring quantified the distortion of a mapping. Lars Ahlfors and Arne Beurling distilled these scattered length-area arguments into a single conformally invariant functional in their 1950 paper Conformal invariants and function-theoretic null-sets ^{[Ahlfors and Beurling 1950]} (Acta Mathematica 83, pp. 101–129), which set extremal length and the associated capacities against the problem of characterising the removable sets for bounded analytic functions. Their paper named the supremum over admissible metrics as the basic invariant and proved that its conformal invariance follows from the length-area exchange under a holomorphic Jacobian, the computation reproduced in the Key theorem above.

The systematic application to open Riemann surfaces is the work of Ahlfors and Leo Sario, whose 1960 monograph Riemann Surfaces ^{[Ahlfors and Sario Riemann Surfaces]} (Princeton University Press) organised the classification theory of non-compact surfaces around extremal length, modulus, and capacity, producing the null-classes $O_G,O_{HB},O_{HD}$ and the parabolicity criterion through divergence of a modulus series. The quasiconformal lineage — Grötzsch 1928, Teichmüller 1939, and the modulus-of-a-quadrilateral distortion theory codified by Olli Lehto and Kalle Virtanen in Quasiconformal Mappings in the Plane ^{[Lehto and Virtanen Quasiconformal Mappings]} (Springer 1973) — established the modulus of a ring and a quadrilateral as the controlling quasiconformal invariant, the quantity whose bounded distortion defines a quasiconformal map. Ahlfors returned to the subject in his 1973 lectures Conformal Invariants ^{[Ahlfors Conformal Invariants]}, which present extremal length as the unifying device of geometric function theory.

Bibliography Master

@article{AhlforsBeurling1950,
  author  = {Ahlfors, Lars and Beurling, Arne},
  title   = {Conformal invariants and function-theoretic null-sets},
  journal = {Acta Mathematica},
  volume  = {83},
  year    = {1950},
  pages   = {101--129}
}

@book{AhlforsSarioRiemannSurfaces,
  author    = {Ahlfors, Lars V. and Sario, Leo},
  title     = {Riemann Surfaces},
  series    = {Princeton Mathematical Series},
  volume    = {26},
  publisher = {Princeton University Press},
  year      = {1960}
}

@book{AhlforsConformalInvariants,
  author    = {Ahlfors, Lars V.},
  title     = {Conformal Invariants: Topics in Geometric Function Theory},
  publisher = {McGraw-Hill},
  year      = {1973}
}

@book{LehtoVirtanen1973,
  author    = {Lehto, Olli and Virtanen, K. I.},
  title     = {Quasiconformal Mappings in the Plane},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {126},
  edition   = {2nd},
  publisher = {Springer},
  year      = {1973}
}

@article{Teichmuller1939,
  author  = {Teichm{\"u}ller, Oswald},
  title   = {Extremale quasikonforme Abbildungen und quadratische Differentiale},
  journal = {Abh. Preuss. Akad. Wiss. Math.-Nat. Kl.},
  volume  = {22},
  year    = {1939},
  pages   = {1--197}
}

@article{Grotzsch1928,
  author  = {Gr{\"o}tzsch, Herbert},
  title   = {{\"U}ber einige Extremalprobleme der konformen Abbildung},
  journal = {Ber. Verh. S{\"a}chs. Akad. Wiss. Leipzig Math.-Phys. Kl.},
  volume  = {80},
  year    = {1928},
  pages   = {367--376}
}

@book{Vaisala1971,
  author    = {V{\"a}is{\"a}l{\"a}, Jussi},
  title     = {Lectures on $n$-dimensional Quasiconformal Mappings},
  series    = {Lecture Notes in Mathematics},
  volume    = {229},
  publisher = {Springer},
  year      = {1971}
}